About WIGTN

AI could do much of what companies asked of it, and then it would fail in ways nobody had measured. A document parser whose output looked fine would quietly hurt retrieval. A voice system would invent words out of silence. An LLM judge would report an improvement that was only noise.

So we worked on those problems, without a lab behind us. This year that work became a system demo at ACL, an industry paper at EMNLP, and a workshop paper at NeurIPS.

Now we take it into enterprise systems, where a wrong answer costs more. We bring the same habit we started with: measure before we claim, and say where the claim stops. If you have a problem AI hasn’t solved for you yet, we’d like to hear about it.